In multi-tenant RAG scenarios, if vector retrieval does not filter by tenant at query time, cleverly worded questions may retrieve documents belonging to other customers, resulting in cross-tenant data leakage.